Google crosses the knife again! How to help NASA without an astronomical background
Do readers remember that INSIDE reported last year that Google AI assisted NASA and discovered the “mini solar system” in which 8 planets in the solar system orbited the sun before. But at the end of last year, astronomers announced that there was a star named “Kepler 90” (Kepler-90) 2545 light years away, with the same number of planets orbiting around it, just like another small solar system.
Breakthrough astronomy research by Google Brain and NASA, previously published in Astronomy, Google Brain research team senior software engineer Chris Shallue analyzed machine learning through video today and shared the research process with us.
Mankind has never stopped thinking and exploring the universe. In 2009, the Kepler mission of NASA used the Kepler telescope to record exoplanets. The slight decrease in the brightness of the star caused by the passing in front of the star. This decrease in brightness can signal Let scientists indirectly calculate the existence of the planet and the physical properties of the planet.
Thirty thousand possible planetary signals have been collected during 4 years of operation. Astronomers use automated analysis techniques with manual checks to interpret light patterns, but the problem is that astronomers can hardly recognize tens of thousands of other signal patterns with unobvious features, but there are still unexplained signals. Astral body, artificial intelligence came in handy, leading to the cooperation between NASA and Google Brain.
In the tens of thousands of signals, many weaker signals also have a lot of noise. It is impossible to manually judge whether the
planet
Applying the previous convolutional neural network that has been quite successful in image recognition, using 15,000 astronomers’ artificially labeled Kepler signals to train the model, and searching for 670 of the Kepler’s database through the model Stars were used to identify new planets, and Kepler-90 i and Kepler-80 g were finally found in a large number of stars.
Maybe you think that machine learning is still difficult to perform in different professions. From the ratio of 670 stars, only two new planets have been found, which is much lower than that recognized by humans, but it is returned to the light map with complex signals. The curve that humans can recognize is obvious, and the machine is responsible for solving massive amounts of data with weak signals. For astronomy and engineers, the cooperation between the two parties has indeed brought breakthrough development, greatly reducing human consumption, and even making it impossible for humans to recognize. The next step is to use the model to calculate more than 200,000 stars in the entire Kepler database.
The Enlightenment of Google’s Multiple Cross-Fields of Machine Learning to Taiwan
In the last year’s machine learning series media gathering, Google shared in sequence how medical, translation, and YouTube video use innovative applications through artificial intelligence. This year’s first machine learning series, they revealed the concept of machine learning in the field of astronomy, and Google Brain’s cross-disciplinary Various projects have successively brought innovative applications to various industries, but returning to the essence of Google Brain, they are from the background of capital workers. What is the purpose of developing new projects to improve machine learning capabilities?
In today’s media gathering, Jian Lifeng specifically mentioned two important points worthy of Taiwan’s attention, the cross-field contribution of engineering and the importance of developing source code. In fact, the research team is engaged in different projects for the purpose of not delving into this field, but hoping to interact with these industries. In the past, it was generally believed that machine learning was a problem for workers, but how to break through the neural network must be integrated with other fields. On the one hand, it can bring the capacity in this field, and on the other hand, it can be better understood through in-depth cooperation. How to make the most effective tools to help people in this field.
Where is the opportunity for Taiwan?
Today, in addition to astronomy, Google also invited another “gene sequencing” project to share cases. Next-generation sequencers provide low-cost and high-throughput tools, but because these DNA reading sequences are quite fragmented, and the results of the sequencer are Many mistakes, so Google Brain worked with the biotechnology team to develop open source tools to improve the accuracy of genome sequencing.
Jian Lifeng mentioned that Taiwan currently has a good foundation in the agricultural industry. In addition, Taiwan’s new venture has invested a lot in genetic testing. If it can continue to integrate machine learning, it will greatly help agriculture, pests and diseases. It can be opened up. Tools make the detection accuracy higher.
When it comes to AI, it immediately reminds you of complex computer computing or esoteric learning model architecture. However, with the development of technology, more and more resources are available to the general public, including the recent addition of AutoML applications to Google Cloud, a cloud service owned by Google, so there is no information. People with backgrounds can build learning models in a simpler way and apply them to its service content. In the future, how our talents will make good use of these resources will be a major turning point for industry breakthroughs.
Further reading:
.